1. Field of the Invention
The invention relates to magnetic memory systems. More particularly, the present invention relates to a method and system for providing an element that employs a spin-transfer effect in switching and that can be used in a magnetic memory, such as a magnetic random access memory (“MRAM”).
2. Description of the Related Art
Magnetic memories are used for storing data. One type of memory element of interest utilizes magnetoresistance properties of a magnetic element for storing data. FIGS. 1A and 1B respectively depict cross sectional views of a conventional spin valve magnetic element 100 and a conventional spin tunneling junction magnetic element 100′. As shown in FIG. 1A, conventional spin valve magnetic element 100 includes a conventional antiferromagnetic (AFM) layer 102, a conventional pinned layer 104, a conventional spacer layer 106, and a conventional free layer 108. Conventional pinned layer 104 and conventional free layer 108 are ferromagnetic. Conventional spacer layer 106 is nonmagnetic and is conductive. AFM layer 102 is used to fix, or pin, the magnetization of the pinned layer 104 in a particular direction. The magnetization of free layer 108 is free to rotate, typically in response to an external field.
Portions of conventional spin tunneling junction magnetic element 100′ are analogous to conventional spin valve magnetic element 100. Accordingly, conventional spin tunneling junction magnetic element 100′ includes an AFM layer 102′, a conventional pinned layer 104′, an insulating barrier layer 106′ and a free layer 108′, as shown in FIG. 1B. Conventional barrier layer 106′is thin enough for electrons to tunnel through.
Depending upon the orientations of the magnetizations of free layer 108 (or 108′) and pinned layer 104 (or 104′), the resistance of conventional magnetic element 100 (or 100) changes. For example, when the magnetizations of free layer 108 and pinned layer 104 are parallel, the resistance of conventional spin valve 100 is low. When the magnetizations of free layer 108 and pinned layer 104 are antiparallel, the resistance of conventional spin valve 100 is high. Similarly, when the magnetizations of free layer 108′ and pinned layer 104′ are parallel, the resistance of conventional spin tunneling junction 100′ is low. When the magnetizations of free layer 108′ and pinned layer 104′ are antiparallel, the resistance of conventional spin tunneling junction 100′ is high.
In order to sense the resistance of either conventional magnetic element 100 or 100′, current is driven through the magnetic element. For conventional magnetic element 100, current can be driven through the magnetic element 100 in one of two configurations: current-in-plane (CIP) and current-perpendicular-to-the-plane (CPP). For conventional spin tunneling junction 100′, however, current is driven through conventional spin tunneling junction 100′using only the CPP configuration. For the CIP configuration, current is driven parallel to the layers of conventional spin valve 100. Thus, as viewed in FIG. 1A, current is driven from left to right or right to left for the CIP configuration. For the CPP configuration, current is driven perpendicular to the layers of conventional magnetic element 100 (or 100′). Accordingly, as viewed in FIG. 1A or 1B, current is driven up or down for the CPP configuration. The CPP configuration is used in MRAM having a conventional spin tunneling junction 100′ in a memory cell.
FIG. 2 depicts a conventional memory array 200 using conventional memory cells 220. Each conventional memory cell 220 includes a conventional magnetic element 100 (or 100′). Conventional array 200 is shown having four conventional memory cells 220, which are typically spin tunneling junction magnetic elements 100′. As shown in FIG. 2, each memory cell 220 includes a conventional spin tunneling junction 100′ and a transistor 222. Memory cells 220 are coupled to a reading/writing column selection 230 via bit lines 232 and 234 and to row selection 250 via word lines 252 and 254. Write lines 260 and 262, which are also depicted in FIG. 2, carry currents that generate external magnetic fields for the corresponding conventional memory cells 220 during writing. Reading/writing column selection 230 is coupled to a write current source 242 and a read current source 240, which are each coupled to a voltage supply Vdd 248 via line 246.
In order to write to conventional memory array 200, a write current Iw 242 is applied to bit line 232 (or 234), which is selected by reading/writing column selection 230. A read current Ir 240 is not applied. Both word lines 252 and 254 are disabled, and transistors 222 in all memory cells 220 are disabled. Additionally, one of the write lines 260 and 262 corresponding to the selected memory cell 220 carries a current used to write to the selected memory cell. The combination of the current in write line 260 (or 262) and the current in bit line 232 (or 234) generates a magnetic field that is large enough to switch the direction of magnetization of free layer 108′ of the selected cell and thus write to the selected cell. Depending upon the data written to the selected cell, the resulting resistance of the conventional magnetic tunneling junction 100′ will be high or low. When reading from a selected cell 220 in array 200, a read current Ir 240 is applied instead. The memory cell selected for reading is determined by row selection 250 and column selection 230. The output voltage of the selected cell is read at output line 244.
Although conventional magnetic memory 200 using conventional spin tunneling junctions 100′ can function, there are obstacles at higher memory cell densities. In particular, conventional memory array 200 is written using an external magnetic field that is generated by currents driven through bit line 232 (or 234) and write line 260 (or 262). That is, the magnetization of free layer 108′ is switched by the external magnetic field that is generated by currents driven through bit line 232 (or 234) and the write line 260 (or 262). The magnetic field required to switch the magnetization of free layer 108′, known as the switching field, is inversely proportional to the width of the conventional magnetic element 100′. As a result, the switching field increases for conventional memories having smaller magnetic elements 100′. Because the switching field is higher for a smaller magnetic element 100′, the required current that is to be driven through bit line 232 (or 234) and, in particular, through write line 260 (or 262) increases dramatically for a higher magnetic memory cell density. This relatively large current can cause a host of problems in a conventional magnetic memory array 200. For example, cross talk and power consumption increases accordingly. Additionally, the driving circuits required for driving the current that generates the switching field at a selected memory cell 220 also increase in physical area and complexity. Furthermore, the conventional write currents must be sufficiently large to switch a magnetic memory cell, but not so large that neighboring cells are inadvertently switched. The upper limit on the magnitude of the write current can lead to reliability issues because cells that are relatively harder to switch than others based on fabrication and material non-uniformity will fail to write consistently.
To overcome some of the obstacles associated with high-density magnetic memories described above, a recently discovered phenomenon, the spin-transfer effect, has been utilized. Details of the spin-transfer effect (and the spin-injection method) are disclosed in J. C. Slonczewski, “Current-driven Excitation of Magnetic Multilayers,” Journal of Magnetism and Magnetic Materials, Vol. 159, pp. L1–L5 (1996); L. Berger, “Emission of Spin Waves by a Magnetic Multilayer Traversed by a Current,” Phys. Rev. B, Vol. 54, p. 9353 (1996), and in F. J. Albert et al., “Spin-polarized Current Switching of a Co Thin Film Nanomagnet,” Appl. Phys. Lett., vol. 77, No. 23, pp. 3809–3811 (2000), each of which is incorporated by reference herein.
The spin-transfer effect offers an alternative switching mechanism that has the advantage of being highly localized. In a multilayer structure consisting of ferromagnetic (or ferrimagnetic) layers that are separated by non-magnetic conducting spacer or an insulating barrier, spin transfer refers to the transferring of spin-angular momentum from the spin-polarized conduction electrons, via quantum-mechanical exchange field, into a ferromagnetic (or ferrimagnetic) layer when an electric current is driven across the multilayers. The spin-transfer effect causes the magnetization vector of the affected ferromagnetic layer to rotate. At high enough current, such a rotation can grow sufficiently large to produce switching (i.e. 180° of rotation). Successful application of the spin-transfer effect to magnetic switching has been demonstrated. Spin-transfer switching has been achieved in a simple giant-magnetoresistance (GMR) spin-transfer structure consisting of Co[thick]/Cu/Co[thin] layers where the Cu layer serves as a spacer separating the two magnetic layers. See, for example, F. J. Albert et al., “Spin-polarized current switching of a Co thin film nanomagnet,” Appl. Phys. Lett., vol. 77, No. 23, pp. 3809–3811 (2000) and J. Grollier et al. “Spin-polarized current induced switching in Co/Cu/Co pillars,” Applied Physics Letters, Vol. 78, No. 23, p. 3663–3665 (2001). If the thickness of the Cu layer is much less than the spin relaxation length, electrons traveling across the Cu layer retain much of their spin polarization. When electrons travel from the thick to the thin Co layer (with the current flowing in the opposite direction), the thick Co layer serves as a spin reservoir that injects spins into the thin Co layer. Transfer of spin-angular momentum causes the magnetization of the thin Co layer to align parallel to that of the thick Co layer.
When electrons travel from the thick to the thin Co layer, on the other hand, electrons having spins that are parallel to the magnetization direction of the thick Co layer pass without much scattering through the thick Co, but electrons having antiparallel spins are reflected. The reflected antiparallel spins cause the magnetization of the thin Co layer to align antiparallel to the magnetization of the thick Co layer. Thus, by changing the flow direction of the spin-polarized current, the magnetization of the thin Co layer can be repeatedly switched to be parallel and antiparallel.
The spin-transfer effect can be used in a CPP configuration as an alternative to or in addition to using an external switching field to switch the direction of magnetization of free layer 108 (or 108′) of a conventional spin valve 100 (or a conventional spin tunneling junction 100′). The spin-transfer effect dominates other mechanisms and thus becomes observable when the dimensions of conventional magnetic element 100 (or 100′) are small, in the range of few hundred nanometers. Consequently, the spin-transfer effect is suitable for higher density magnetic memories having smaller magnetic elements 100 (or 100′).
For example, switching the magnetization of a conventional free layer 108 in conventional spin valve 100 using spin transfer can be carried out as follows. Current can be driven from conventional free layer 108 to conventional pinned layer 104 to switch the magnetization of conventional free layer 108 so that it is parallel to the magnetization of conventional pinned layer 104. The magnetization of conventional free layer 108 is assumed to be initially antiparallel to conventional pinned layer 104. When current is driven from conventional free layer 108 to conventional pinned layer 104, conduction electrons travel from conventional pinned layer 104 to conventional free layer 108. The majority electrons traveling from conventional pinned layer 104 have their spins polarized in the same direction as the magnetization of conventional pinned layer 104 and interact with the magnetic moments of conventional free layer 108 near the interface between conventional free layer 108 and conventional spacer layer 106. As a result, the electrons transfer their spin-angular momentum to conventional free layer 108. Thus, angular momentum corresponding to spins that are antiparallel to the magnetization of conventional free layer 108 (and parallel to conventional pinned layer 104) is transferred to the conventional free layer. If the electrons transfer sufficient angular momentum, the magnetization of conventional free layer 108 is switched to be parallel to the magnetization of conventional free layer 104.
Alternatively, current can be driven from conventional pinned layer 104 to conventional free layer 108 to switch the magnetization of conventional free layer 108 to be antiparallel to the magnetization of conventional pinned layer 104. In this situation, the magnetization of free layer 108 is assumed to be initially parallel to the magnetization of pinned layer 104. When current is driven from conventional pinned layer 104 to conventional free layer 108, conduction electrons travel in the opposite direction. The majority electrons have their spins polarized in the direction of magnetization of conventional free layer 108, which is originally magnetized in the same direction as conventional pinned layer 104. The majority electrons are transmitted through conventional pinned layer 104. The minority electrons, however, which have spins that are polarized antiparallel to the magnetization of conventional free layer 108 and conventional pinned layer 104, will be reflected from conventional pinned layer 104 and travel back to conventional free layer 108. The minority electrons reflected by conventional pinned layer 104 interact with magnetic moments of conventional free layer 108 and transfer at least a portion of their spin-angular momentum to conventional free layer 108. If the electrons transfer sufficient angular momentum to conventional free layer 108, the magnetization of free layer 108 is switched to be antiparallel to the magnetization of conventional pinned layer 104.
In the CPP configuration, therefore, a current driven through a conventional magnetic element 100 (or 100′) can switch the direction of magnetization of free layer 108 (or 108′) using the spin-transfer effect. That is, the spin-transfer effect can be used to write to magnetic element 100 (or 100′) in a magnetic memory by driving a current through a conventional magnetic element 100 (or 100′) in a CPP configuration. Thus, writing using the spin-transfer effect is more localized and, consequently, generates less cross talk. Spin-transfer effect is also more reliable because spin-transfer results in a high effective field in a conventional magnetic element 100 (or 100′) in a device, such as MRAM. Additionally, for a magnetic element 100 (or 10′) having a small enough size, the current required to switch the magnetization using the spin-transfer effect can be significantly less than the current required to generate a switching field in conventional magnetic memory array 200. Accordingly, there is less power consumption when writing.
Although the spin-transfer effect can be advantageously used to switch the direction of the magnetization of conventional free layer 108 (or 108′), there are barriers to using the spin-transfer effect for a conventional magnetic element 100 (or 100′) in a memory. For a conventional spin valve magnetic element 100, the CPP configuration results in a significantly reduced signal. For example, the magnetoresistance ratio for the CPP configuration of a conventional spin valve 100 is only approximately two percent. Moreover, the total resistance of a conventional spin valve magnetic element 100 is low. Thus, the read signal output by a conventional spin valve magnetic element 100 is very low. Consequently, although the spin-transfer effect can be used to write to a conventional spin valve magnetic element 100, the output signal when reading from the conventional spin valve 100 is so low that it is difficult to use a conventional spin valve magnetic element 100 in a magnetic memory that is written using the spin-transfer effect.
On the other hand, a conventional spin tunneling junction magnetic element 100′ has a large resistance-area product (Ra) that is typically on the order of kΩ μm2. The high current density that is required for inducing the spin-transfer effect, however, could destroy the thin insulating barrier due to ohmic dissipation. Moreover, the spin-transfer effect has not been observed in a conventional spin tunneling junction magnetic element 100′ at room temperature. Thus, an MRAM having conventional spin tunneling junction magnetic elements 100′ that have a high Ra value may not be able to be used with the spin-transfer effect for writing to the magnetic memory cells.
As used herein, the term “spin-transfer stack” is defined to be any two-terminal multilayer structure that utilizes the spin-transfer effect for switching the magnetization of a ferromagnetic (ferrimagnetic or sperimagnetic) layer. The term “two-terminal multilayer structure (or device)”, as used herein, is defined to be a multilayer structure (or device) that has one bottom lead in contact with a bottom layer of a multilayer structure and one top lead in contact with a top layer of the multilayer structure, and with current driven between the bottom and top lead through the multilayer structure. In contrast, U.S. Pat. No. 5,695,864 to Slonczewski discloses a three-terminal device using spin-transfer for switching. The structure and operational principles of such a three terminal device are different from the spin-transfer stack devices described herein. All spin-transfer stack devices referred to herein will be assumed to be two-terminal, unless otherwise noted.
To date, several types of two-terminal spin-transfer stack devices have been proposed. A first type of two-terminal spin-transfer stack device is referred to herein as a single-spacer spin-transfer stack device. A second type of two-terminal spin-transfer stack device is referred to herein as a single-barrier & dual barrier spin-transfer stack device. A third type of two-terminal spin-transfer stack device is referred to herein as a spacer-barrier spin-transfer stack device.
An exemplary single-spacer spin-transfer stack device is formed having a fixed magnetic layer/Cu spacer/Free magnetic layer. Variations of a single-spacer spin-transfer stack device include additions of antiferromagnetic layers for pinning and synthetic layers (such as CoFe[20A]/Ru[8.4A]/CoFe[22A]) for reducing magnetostatic coupling between the layers. The single-spacer spin-transfer stack type suffers numerous disadvantages, including an extremely small signal, a low resistance, and a high switching current.
An exemplary single-barrier stack device is formed having a Fixed magnetic layer/Tunnel barrier/Free magnetic nanoparticle. See, for example, U.S. Pat. No. 6,256,223 to Sun. An exemplary dual-barrier stack device is formed having a Fixed magnetic layer/Tunnel barrier/Free magnetic nanoparticle/Tunnel barrier/Fixed magnetic layer. See, for example, U.S. Pat. No. 6,256,223 to Sun. Variations to this type of two-terminal spin-transfer stack device include additions of antiferromagnetic layers for pinning and synthetic layers for reducing magnetostatic coupling between the layers. The single-barrier and dual-barrier spin-transfer stack devices suffer from numerous disadvantages, including an extremely high resistance, a high switching current, an unreliable method of fabrication, and a lack of working experimental evidence of spin-transfer effect switching across tunnel barrier(s) at room temperature.
An exemplary spacer-barrier stack device is formed having a Fixed magnetic layer/Conducting spacer/Free magnetic layer/Low-resistance tunnel barrier/Fixed magnetic layer. See, for example, copending and co-assigned U.S. patent application Ser. No. 10/213,537, filed Aug. 6, 2002, entitled Off-Axis Pinned Layer Magnetic Element Utilizing Spin Transfer And An MRAM Device Using The Magnetic Element, which is incorporated by reference herein. Variations to a spacer-barrier spin-transfer stack device include additions of antiferromagnetic layers for pinning and synthetic layers for reducing magnetostatic coupling between the layers. The spacer-barrier stack types address most of the problems of the other spin-transfer stack types. The low-resistance tunnel barrier significantly increases the read signal, but yet keeps the total stack resistance low. Unlike spin-transfer switching across a tunnel barrier, spin-transfer switching across a conducting spacer at room temperature has been experimentally observed. Consequently, a spacer-barrier spin-transfer stack is guaranteed to switch because it contains a conducting spacer arrangement.
Giant magnetoresistive (GMR) and tunnel magnetoresistive (TMR) ratios are strongly dependent on the spin polarization P of the magnetic layers. Spin polarization P can be defined straightforwardly as the percentage of up (down) spins minus the percentage of down (up) spins in a ferromagnetic material at the Fermi level. Most common magnetic materials, such as Fe, Co, Ni and their alloys have a rather low P of less than 50%. Half-metals are ferromagnets having a very high P (approaching 100%) that are metallic in one spin orientation and insulating in the other spin orientation. Because of their potential application to GMR and TMR devices, half-metals recently have been actively researched. Some half-metallic materials that have been discovered are CrO2, Sr2FeMoO6, (La0.7Sr0.3)MnO3, Fe3O4, and NiMnSb, of which Fe3O4 has the highest Curie temperature resulting in very high spin polarization P at room temperature.
The low spin-polarization materials have a significant number of minority spins that have an orientation that is opposite to that of the majority spins, despite a display of permanent magnetization. When these low spin-polarization materials are used in magnetic layers in a spin-transfer stack, the spin-transfer effect is reduced as the transferred angular momentum from the minority spins negates much of the transferred angular momentum from the majority spins. As a result, the switching current must be large to induce switching. On the other hand, when high spin-polarization half-metallic materials are used as magnetic layers in a spin-transfer stack, the opposing angular momentum negation from the small (theoretically zero for a perfect half-metal) number of minority spins is also small, resulting in a small switching current requirement. The decrease in the switching current can be significant, i.e., by more than a factor of 5, when going from materials of spin polarization of 35% to materials of spin polarization of 80% assuming coherent electron scattering in the free magnetic layer. See, for example, J. C. Slonczewski, “Current-driven excitation of magnetic multilayers,” Journal of Magnetism and Magnetic Materials, Vol. 159, L1–L7 (1995).
Previously, half-metals have been introduced, at least conceptually, for improving performance of spin valves and spin tunneling junctions. Typical improvement goals for the introduction of half-metals have been higher GMR and TMR ratios. Half-metals can be introduced in various forms, such as a total replacement of all non-half-metallic ferromagnetic layers, a partial replacement, a thin coating at the appropriate interfaces between the layers, and as thin film insertion within ferromagnetic layers. Introduction of half-metals to spin valves and spin tunneling junctions, however, provide devices that do not involve spin-transfer switching or writing.
Partial introduction of half-metals to single-spacer spin-transfer stack has also been proposed previously. U.S. Pat. No. 5,696,864 to Slonczewski discloses that all of the pinned and/or free magnetic layers in a single-spacer spin-transfer stack device can be replaced with high spin-polarization or half-metallic magnetic materials. Replacing a magnetic layer with a high spin-polarization of half-metallic material, though, can lead to problems related to the material characteristics of the new material. For instance, if the half-metallic material, such as Fe3O4, has extremely high resistivity, the total resistance of the spin-transfer stack would be extremely high, thereby resulting in a very small signal. The half-metallic material may also have large surface roughness that would affect the other layers in the spin-transfer stacks. Finally, if the half-metallic material has high magnetic anisotropy and coercivities (e.g., greater than 1000 Oe), the half-metallic material cannot be practically used for replacing the free layer.
Nevertheless, all of the spin-transfer stacks described above suffer from a high switching current. Consequently, what is needed is a way to reduce the high switching current exhibited by a conventional spin-transfer stack. Additionally, what is still needed is a magnetic memory element that can be used in a memory array having high density, that provides low power consumption, low cross talk, and has high reliability, while providing a useable read signal.